Subjective cognitive decline predicts incidence of mild cognitive impairment and dementia among older adults without cognitive impairment (Mitchell et al., 2014), suggesting self-reported cognitive decline may be a sensitive measure of cognitive change. However, most of this research has been conducted among highly educated non-Latino white participants recruited from memory clinics, and little is known about social factors that modify the correspondence between objectively measured cognitive function and subjective cognitive decline. Prior work has shown that anxiety and depressive symptoms influence the correspondence between objectively measured cognitive function and subjective cognitive decline such that people with higher levels of anxiety or depressive symptoms are more likely to report subjective cognitive decline for the same level of objectively measured cognitive function (Hanninen et al. 1991; Schmand et al. 1997; Reid et al. 2006, reviewed in Jonker et al. 2000). However, to our knowledge, no prior work has evaluated whether social factors modify the link between objectively measured cognitive function and subjective cognitive decline. For example, people with a family history of dementia may be more aware or concerned about subtle changes in cognitive function, which may result in people with a family history of dementia to report more subjective cognitive decline for at a given level of objectively measured cognitive function.
Does the correspondence between objectively measured of cognitive function (SENAS) and self-reported decline in cognitively-related functional ability (ECog) differ by race/ethnicity, gender, educational attainment, family history of dementia, and depressive symptoms(?) in a diverse sample of older adults without diagnosis of dementia?
The dataset consisted of a Cross-sectional analysis of the baseline data from the Kaiser Healthy Aging and Diverse Life Experiences KHANDLE study (Kaiser-Permanente Health AND Life Exposures)
Some variables have been recoded or created as follows: EDUCATION consists of a categorical variable specific to post-high school and doesn’t account for vocational diploma and trade school. EDUCATION also doesn’t separate obtained high school diploma (or equivalents) from uncompleted high school.
The variable TRNCERT indicates if the participant obtained a certificate:
and the variable LONGCERT indicates how long it took (values between 1 and 4 with 4 = “6 months or more”)
we created a variable (TRUE_CERT) to indicate whether a certificate respects both conditions (TRNCERT =2 and LONGCERT=4) if both conditions are true the number of years of education is coded as the actual number of years of education (contained in EDUCATION_TEXT whenever education is 12 years or less) + 1 (only for participants that have 12 years or less)
For the other education levels (with college attendance) coded in the variable EDUCATION as integers between 0 (=no college) to 5 (=PHD or equivalent), the Education will be recoded as a continuous variable (yrEDUCATION) as follows (see table):
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 16 | 18 | 20 | NA | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0: no college | 4 | 1 | 3 | 5 | 4 | 3 | 13 | 13 | 12 | 22 | 30 | 25 | 152 | 46 | 0 | 0 | 0 | 0 | 0 |
| 1: some college no dgr | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 319 | 0 | 0 | 0 | 0 | 0 |
| 2: Associate | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 177 | 0 | 0 | 0 | 0 |
| 3: Bachelor | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 410 | 0 | 0 | 0 |
| 4: Master | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 275 | 0 | 0 |
| 5: PhD or equiv. | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 102 | 0 |
| missing | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Number of PARENTS with dementia (PARENTAL_DEMENTIA) | |
| 0 | 1170 |
| 1 | 414 |
| 2 | 32 |
| Number of siblings with dementia (SIBLING_DEMENTIA) | |
| 0 | 1482 |
| 1 | 113 |
| 2 | 17 |
| 3 | 2 |
| 4 | 1 |
| 5 | 1 |
| Has any relative with dementia (RELATIVE_DEMENTIA) | |
| no | 1077 |
| yes | 539 |
|
Total No. 1,616 |
|
|---|---|
| Age | 75.9 (6.7) |
| No. age 90+ | 70 (4.3) |
| Race/Ethnicity | |
| Asian | 400 (24.8) |
| Black | 409 (25.3) |
| Latino | 330 (20.4) |
| Non-Latino-White | 477 (29.5) |
| Gender (Women) | 956 (59.2) |
| Years of Education | 14.7 (3.1) |
| Depressive symptoms | -0.1 (0.7) |
| Family history of dementia | 539 (33.4) |
| Episodic memory score | 0.0 (1.0) |
| Executve function score | 0.0 (1.0) |
| ECog score | 1.4 (0.4) |
| log(ECog) score | 0.3 (0.3) |
ECog_avg = b0 + b1:SENAS + b2:Age + b3:Language_of_interview + b4:Race/ethnicity + b5:Gender + b6:Education + b7:Family_history + b8:Depressive_sx + b9:Race/ethnicity:SENAS + b10:Gender:SENAS + b11:Education:SENAS + b12:Family_history:SENAS + b13:Depressive_sx:SENAS + b14:Age:SENAS
The everyday cognition scale (Ecog) Farias et al., 2008 was initially informant-based but later used both as an informat-based and self-reported measure of subjective cogntive function either cross-sectionally or longitudinally Farias et al., 2009a; 2009b; 2010. The later developped short version for informant-based assesments of everyday cognition contains 12 items with good internal consistency and efficiently discriminated between cognitively healthy participants and participants with mild or advanced cognitive impairement Faris et al. 2011. However, the short version the Ecog has not been used in self-reported evaluations so far.
The items included in the short version are comprized of questions evaluating if participants are capable of:
| Detail of the Ecog scale | |
|
Total No. 1,616 |
|
|---|---|
| ECOG_MEM1 | |
| Mean (SD) | 2.0 (0.9) |
| Missing | 9 (0.6) |
| ECOG_MEM2 | |
| Mean (SD) | 1.5 (0.7) |
| Missing | 11 (0.7) |
| ECOG_LANG1 | |
| Mean (SD) | 1.6 (0.8) |
| Missing | 6 (0.4) |
| ECOG_LANG2 | |
| Mean (SD) | 1.5 (0.8) |
| Missing | 15 (0.9) |
| ECOG_VISUAL_SPATIAL2 | |
| Mean (SD) | 1.3 (0.6) |
| Missing | 84 (5.2) |
| ECOG_VISUAL_SPATIAL5 | |
| Mean (SD) | 1.1 (0.3) |
| Missing | 6 (0.4) |
| ECOG_PLANNING1 | |
| Mean (SD) | 1.1 (0.4) |
| Missing | 17 (1.1) |
| ECOG_PLANNING3 | |
| Mean (SD) | 1.2 (0.5) |
| Missing | 16 (1.0) |
| ECOG_ORGANIZATION1 | |
| Mean (SD) | 1.5 (0.8) |
| Missing | 8 (0.5) |
| ECOG_ORGANIZATION2 | |
| Mean (SD) | 1.2 (0.5) |
| Missing | 106 (6.6) |
| ECOG_DIVIDED_ATTENTION1 | |
| Mean (SD) | 1.5 (0.8) |
| Missing | 25 (1.5) |
| ECOG_DIVIDED_ATTENTION2 | |
| Mean (SD) | 1.3 (0.7) |
| Missing | 38 (2.4) |
| Ecog12_missing | 0.2 (0.6) |
| Ecog12_including_partial_averages | 1.4 (0.4) |
Figure 6.1: Distribution of ECOG values(including incomplete cases)
Note that the distribution is still strongly skewed even for the standardized log.
The Model as stated in 5.3 assumes linear relationships between SENAS and Ecog. To get a litle more confidence in this, Maria G recommended that we investigate this assumption.
The three following plots show rcs regressions with hinges at the Frank Harrell quantiles (5, 27.5, 50, 72.5 and 95 %)
Figure 7.1: Relationship between Ecog and SENAS cognitive scores
In the three groups of plots above, the relationship between between Ecog and executive function from the SENAS seems to have different slopes in the low values (<1) and the hight values (>1) So in the following analysis we will include two linear splines for this variable.
note: the knot at executive function = 1 is still relevant after transforming Ecog into log(Ecog)
We will sequentially analyse the consequtive models as follows: Our main model always includes Age, as this variable is known to alway be associated with both, our exposure variables and our outcome variable.
Episodic memory:
Ecog_avg = memory + AGE + language + memory*AGE
Executive function:
Ecog_avg = ex_fun<1 + ex_fun>1 + AGE + language + ex_fun<1*AGE + ex_fun>1*AGE
Figure 8.1: histogram of residuals and plot of predicted values
We constructed consecutive models In the following order : * “age + race” * “age + gender” * “age + years of education + race” (because edu is patterned by race) * “age + family dementia” * “age + depressive symptoms”
a link to a formatted output here
## png
## 2
In addition to the present full model, we could add multiple interaction between the different regessors (like age:education:gender)
A link to a formatted output here:
## png
## 2
## png
## 2
## png
## 2
## png
## 2
## png
## 2
## png
## 2
## png
## 2